Clinical translation of fluid, imaging, and digital biomarkers for Alzheimer’s disease

Abstract

<p>Background: The Centiloid (CL) scale was developed to harmonise the quantification of amyloid-beta (Ab) positron emission tomography (PET) and is currently widely implemented in Alzheimer’s research. However, this scale could remain vulnerable to effects beyond image analysis.</p><p>Aim: The purpose of the present study was to determine the reliability of the transformation of [11C]PiB PET data acquired at Turku PET Centre to the CL scale.</p><p>Method: First-level calibration anchor points were calculated following the CL standard method with 45 Alzheimer’s disease patients and 34 young controls from the Global Alzheimer’s Association Interactive Network (GAAIN) dataset. The GAAIN dataset was re-processed with Turku PET Centre’s pipeline Magia (Karjalainen et al., 2020) for level-2 calibration of a non-standard image analysis method. Standardized uptake value ratios (SUVRs) were referenced to the whole cerebellum (WC) and the corrected cerebellar grey matter (cCG). The direct equations for CL conversion were applied to 101 [11C]PiB PET scans from cognitively healthy older adults (mean age 71.3 years, 59.4% females, 60.4% APOE4 carriers) acquired with a High-Resolution Research Tomograph. A data-driven threshold for Ab positivity was extracted from each method with a gaussian mixture model.</p><p>Result: First and second level analyses with the calibration dataset yielded a slope and correlation coefficient near unity (y=0.999x+0.075, R2=0.997, and y=1.01x+0.096, R2=0.961, respectively). Values referenced to WC resulted in a threshold of 25.43 CL, in accordance with the proposed CL threshold of 24–30 CL. However, direct conversion of SUVRs referenced to cCG resulted in higher CLs, compared to WC reference region, with a data-driven threshold of 41.18 CL for Ab positivity. <br></p><p>Conclusions: Transformation of [11C]PiB PET data processed with non-standard pipelines to CLs is possible, but affected by the choice of reference region. SUVRs referenced to WC appear to be more generalisable.</p>